**Row 163 r2 review (Filbert): approve** **What I reviewed:** - Candidate 2c090eefd on code 2a5f3a3a0, base 4f02486d5. - Manifest c6f4f4c12596: all 10 files match (`sha256sum -c`). - r2/candidate.diff in packet 7af166574 is byte-identical to `git diff 4f02486d5 2c090eefd`. - Base drift: nothing has landed on these files on refactor since the base. - Since r1 (4be3f6aec), four files changed: stack-release.test.mjs, session.mjs, session.test.mjs and BUILD.md. stack-release.sh is still blob 0053f35d1360; cli.mjs, launcher.mjs and TOOLS.md are unchanged. **Tests I ran** (my own worktree, 15:26-15:39Z, TMPDIR ~/fs1632, totals summed by script): - stack-release.test.mjs: 166/166 - launcher.test.mjs: 29/29 - host.test.mjs: 42/42 - session.test.mjs: 12/12 - Total: 249 pass, 0 fail. **Mutants:** I reran my r1 mutants and added N1. All seven are now killed: | Mutant | What it breaks | Killed by | | --- | --- | --- | | M1 | `live-config-unchanged` always passes | R14 | | M2 | `backup-unchanged` always passes | R14 | | M3 | `credential-ref-live` accepts an env ref | R9 | | M4 | `data-root-scratch` mismatch dropped | R11 | | M5 | `no-session` ignores `launches/` | R4 | | M6 | no INT/TERM/HUP trap | R15 | | N1 | read-only return in `Registry #write()` dropped | session test "a read-only Registry never writes sessions.json, refuses add() and leaves the file on remove()" | **My r1 findings:** - **F1, fixed.** R14 uses a host-touches fault and covers three files: the live config, the live business file and the backup bus.sqlite. Each case gives exactly one FAIL by name, after PASS gate and PASS host-stopped. - **F2, fixed.** - R9: an env case with the exact message. - R11: an other-root case and a no-root case, both refused before the host starts. - R4: a host-launches-dir case. - **F3, fixed.** R15 sends TERM, INT and HUP while the gate is waiting on /healthz. - It asserts the full name list: PRE plus migration, FAIL signal, then all five stop and end checks PASS. - It also checks REHEARSAL_FAIL, exit 1, and that the host pid is gone. - The Console's stop is proven by PASS console-stopped. - **N1, fixed.** `#write()` returns at once under read-only, so the constructor, add() and remove() all follow the same rule, and the session test covers remove(). **Read and found correct:** - The new rHost faults only act when their fault file is present, so the existing cases don't change. - The `e.during` hook is optional (`?.`), so the other R tests don't change. - The session.mjs header comment matches the new behavior. BUILD.md's round 2 section matches the code. - No test pins a hash of any changed file. **Landing:** per Mos's order, this lands after B7, with a fresh gate if the base moves.